September 2022
From: Brian and Tobias
Subject: Cost cutting, abstraction, and chips
Happy Fall! Tobias and I are back with our September edition of our infrastructure newsletter. For those of you who are new to this distribution, this is where we share trends spotted and opportunities to get involved in the work we’re doing at Primary.
Coming out of the summer, deal pacing has picked up and we’re once again in a groove of meeting tons of founders. There are three themes emerging from conversations with folks in and around the industry that we’re further exploring.
Cost Cutting:
In our last newsletter, we wrote about the pattern of “bundling and unbundling” unfolding in developer tools and infrastructure. We have companies in our portfolio playing on this theme across open source infrastructure, security, and MLOps. Bundling can happen in a market when there are many disparate solutions that should exist under one company. Offering these solutions in a fragmented way is more expensive than housing them in one place.
An adjacent theme is cost cutting and optimization. We are seeing a big wave of cost optimization startups in infrastructure. In the last month or so alone, we have looked at several startups attempting to optimize cloud spend, and Bluesky publicly launched to offer this service for data warehouses (specifically Snowflake to start). Recently, we have gotten interested in the ascension of Cribl, a tool for optimizing spend across multiple observability tools (or alternatively, a “data lake for observability”). We have been researching opportunities to build something similar for a different part of the infra stack – are there other examples of pieces of infrastructure where companies “double up” on tooling, resulting in unnecessary spend and fragmented data?
Abstraction:
Another theme that we’re keyed into is abstraction, which can mean either (1) removing some layer of work by automating it away or making it simpler, or (2) giving someone the superpowers to perform another job or function by translating that job into their native language. These two ideas become blurry at times but both enable job abstraction and accessibility.
In part, this was the genius of generational infrastructure projects like Kubernetes, Terraform, and dbt. In fact, the tagline on the front of the dbt website reads, “dbt helps data teams work like software engineers.” Supercharging data engineers with the skills and capabilities of a software engineer is extremely powerful. That’s an interesting framework - where else can it be applied?
Entrepreneurs are pulling at this thread in a whole bunch of ways. We recently met with a company whose pitch was turning software engineers into ML engineers. We know a different founder looking to build a product that abstracts away data engineers completely. Time will tell which abstractions actually work and result in real enterprise value creation, but it’s a theme we are actively exploring.
Chips:
There is a lot of interesting activity happening in the semiconductor industry. Geopolitical tensions in East Asia, combined with the CHIPS Act in the US, means that there will be onshoring of semiconductor design and manufacturing to come. In the same way that the US previously made a push to become energy independent for sovereignty and strategic reasons, the US could also seek to become “chips independent” in the years to come.
Additionally, technological innovation in the semiconductor industry is exploding. Advancing AI and ML has increased the complexity required for chips. Investment in semiconductor companies reached a record $20B in 2021, despite long being a wasteland of venture capital dollars. NVIDIA recently announced its omniverse cloud offering, its first SaaS offering to enable the creation of metaverse applications. The offering is enabled by huge breakthroughs in chip design, with an architecture codenamed “Ada Lovelace.” Ben Thompson has written at length about NVIDIA and these innovations, and his posts on the topic are worth a read.
Almost all of the recent VC funding has gone towards innovative semiconductor companies. Very little, if any, has gone towards software and infrastructure to enable the growth of the semiconductor industry domestically. Whether that’s an actual opportunity or not is unclear, but we’re interested in finding out. We’re currently talking to a company called DFiant trying to introduce a new framework and language for chip design, which is an interesting angle with all sorts of commercial and GTM question marks but lots of potential. If folks have friends or colleagues interested in this market, we’d love to connect.
If you have feedback on these emails or would like to see something else included going forward, please let us know.
Until next time,
Brian and Tobias